# Degrees of freedom of classical ﬁelds

Sergei Winitzki
(Dated: DRAFT December 3, 2006)
This is a pedagogical account of the concept of “independent degrees of freedom” for classical
ﬁelds. I propose two deﬁnitions and show how to compute the number of degrees of freedom for the
electromagnetic and the gravitational ﬁelds.
I. DEGREES OF FREEDOM IN MECHANICS
AND IN FIELD THEORY
In classical mechanics, one says that a system has n
degrees of freedom if it can be unambiguously described
by specifying n real-valued functions of time q
1
(t), ...,
q
n
(t), called generalized coordinates, and if it cannot be
described by a fewer number of coordinates. (Of course,
a system can always be described by a larger number of
coordinates than strictly necessary.) For example, two
point masses m
1
, m
2
moving in three-dimensional space
and connected by a rigid, inﬁnitely thin massless rod
comprise a mechanical system with ﬁve degrees of free-
dom. The state of this system can be speciﬁed by the
values of three Euclidean coordinates x
1
, y
1
, z
1
of the
ﬁrst point mass and two spherical angles φ, θ describing
the orientation of the rod.
To compute the number of degrees of freedom (DOF)
in a mechanical system, one starts with a description of
the system in some set of generalized coordinates, looks
at the equations of motion, and determines the number
of independent integration constants needed to describe
the general solution. This number is equal to twice the
number of the DOF. (The factor 2 comes from having co-
ordinates q
i
and velocities ˙ q
i
for each DOF.) For example,
the equations of motion
¨ x = y, ¨ y = 1 (1)
allow four independent constants of integration, and
therefore the system has two degrees of freedom.
The computation of the DOF in mechanics is relatively
simple because every system is made of a ﬁnite number of
point masses. In ﬁeld theory, the basic object of descrip-
tion is not a point mass but a ﬁeld which is a function of
space and time. For example, the state of a scalar ﬁeld
(e.g. charge density) is described by a function ρ(t, x),
while an electromagnetic ﬁeld can be described by the
4-vector potential A
µ
(t, x). According to the deﬁnition
used in mechanics, any ﬁeld has inﬁnitely many degrees
of freedom since one needs inﬁnitely many functions of
time (such as A
µ
(t, x
1
), A
µ
(t, x
2
), etc., for every possible
vector x) to describe the states of the system. One would
like to have a more discriminating deﬁnition of the DOF,
so that diﬀerent ﬁelds can be distinguished by their num-
ber of DOF. For instance, we would like to say that the
scalar ﬁeld ρ(t, x) has “fewer” DOF than the vector ﬁeld
A
µ
.
Indeed, the deﬁnition of DOF in ﬁeld theory is diﬀerent
from that in ordinary mechanics. We shall ﬁrst consider
a straightforward deﬁnition of DOF and later consider a
diﬀerent meaning of “DOF” and give a second deﬁnition.
Deﬁnition 1: A ﬁeld has n DOF if its physical states
can be unambiguously described by n real-valued func-
tions of space and time, and no fewer number of functions
of space and time can suﬃce.
According to this deﬁnition, a scalar ﬁeld has one
DOF, while a vector ﬁeld, such as A
µ
, has four DOF
because the four components A
0
, A
1
, A
2
, A
3
are real-
valued functions. However, this conclusion is not quite
correct for the electromagnetic ﬁeld because of the gauge
invariance of electrodynamics. A gauge transformation
changes A
µ
(t, x) to another function,
A
µ
(t, x) →
˜
A
µ
(t, x) = A
µ
(t, x) + ∂
µ
α(t, x), (2)
where α(t, x) is an arbitrary function of space and time.
However, the physically observable eﬀects of the ﬁelds
A
µ
and
˜
A
µ
are the same, i.e. we cannot distinguish be-
tween the states of the electromagnetic ﬁeld described by
A
µ
and
˜
A
µ
. In other words, the two diﬀerent functions
A
µ
(t, x) and
˜
A
µ
(t, x) describe the same physical state of
the electromagnetic ﬁeld. Since the gauge freedom con-
tains an arbitrary function of space and time, we expect
that the electromagnetic ﬁeld actually has fewer than four
DOF.
In the particular case of the gauge transformation (2),
it is easy to see that the function α can be chosen to
set A
0
(t, x) = 0. The remaining three functions form a
3-vector potential A ≡ (A
1
, A
2
, A
3
) and are still subject
to the residual gauge freedom,
A(t, x) →
˜
A(t, x) = A(t, x) +∇α(x). (3)
However, the function α now may depend only on x (but
not on t). We only count functions of all four coordinates,
and thus the residual gauge transformations (3) do not
reduce the eﬀective number of the DOF. We conclude
that the electromagnetic ﬁeld has three DOF. In Deﬁni-
tion 1, the phrase “physical states” must be understood
as “ﬁeld conﬁgurations up to gauge freedom.”
II. DEGREES OF FREEDOM IN VACUUM
In some situations, it is necessary to consider a more
restrictive deﬁnition of the number of DOF. In partic-
ular, care should be taken when dealing with ﬁelds in
backgrounds or those having higher-order equations of
motion.
2
An important issue is to describe vacuum solutions,
i.e. nontrivial solutions in the absence of sources. For
example, electromagnetic waves are vacuum solutions of
electrodynamics because these solutions can propagate in
vacuum, in the absence of any charges or currents. We ex-
pect that in a given theory there are inﬁnitely many vac-
uum solutions and that all these solutions are described
by a certain number of arbitrary functions. This num-
ber is what we call the number of vacuum DOF. We shall
ﬁrst count the number of vacuum DOF for electrodynam-
ics and then formulate a more general deﬁnition which is
applicable also to nonlinear ﬁeld theories such as general
relativity.
A. Vacuum DOF in electrodynamics
The 4-vector potential A
µ
(t, x) is a solution of the
Maxwell equations which may be written as

ν

ν
A
µ
−∂
µ
(∂
ν
A
ν
) = j
µ
, (4)
where j
µ
(t, x) is the 4-vector describing the sources
(charges and currents). Note that the current j
µ
must
satisfy the conservation law

µ
j
µ
= 0, (5)
which follows from Eq. (4) by applying ∂
µ
to both sides; if
the current j
µ
is not conserved, Eq. (4) cannot be solved.
We shall now consider the vacuum situation, j
µ
= 0.
An example of a vacuum solution is the plane wave
A
µ
(t, x) = a
µ
cos (k
ν
x
ν
) , (6)
where a
ν
is a 4-vector describing the amplitude of the
wave and k
ν
is the wave vector. Let us ﬁnd the number
of independent solutions of the form (6). By substitut-
ing into Eq. (4) with j
µ
= 0 one can verify that the
ansatz (6) is a vacuum solution only if k
ν
k
ν
= 0 and
a
ν
k
ν
= 0, i.e. k
µ
must be a null 4-vector and a
µ
must
be orthogonal to k
µ
. All 4-vectors a
µ
orthogonal to k
µ
form a three-dimensional subspace spanned by k
µ
itself
and two spacelike vectors that we shall denote l
µ
and m
µ
.
To ﬁnd the components of the vectors l
µ
and m
µ
explic-
itly, consider the components of the vector k
µ
≡ (k
0
, k)
and choose any pair of 3-vectors l, m which are orthogo-
nal to each other and to the 3-vector k; then the vectors
l
µ
≡ (0, l) and m
µ
≡ (0, m) are orthogonal to each other
and to k
µ
. Therefore the admissible vacuum solutions of
the form (6) must have the amplitude vector a
µ
of the
form
a
µ
= αk
µ
+ βl
µ
+ γm
µ
, (7)
where α, β, γ are constant coeﬃcients (which are so far
arbitrary). Furthermore, a gauge transformation of the
form
A
µ
→A
µ
+ ∂
µ
[λsin (k
ν
x
ν
)] (8)
changes the amplitude by a
µ
→ a
µ
+ λk
µ
, where λ is
an arbitrary constant. Now the term αk
µ
in Eq. (7) can
be removed by a gauge transformation (8) with λ = −α,
and the remaining freedom in choosing the amplitude a
µ
consists of choosing the two coeﬃcients β, γ. Thus the
admissible plane waves of the form (6) are described by
just two free parameters.
We are getting near the (correct) conclusion that the
vacuum solutions of the Maxwell equations have only two
DOF. However, this conclusion is not yet fully justiﬁed
because the plane waves of the form (6) are not the only
possible vacuum solutions. Let us therefore imagine the
set of all vacuum solutions and try to analyze the struc-
ture of this set.
Since Eq. (4) with j
µ
= 0 is linear in A
µ
and homo-
geneous, all vacuum solutions form a linear space (linear
combinations of vacuum solutions are again vacuum so-
lutions). This linear space is inﬁnite-dimensional because
there are inﬁnitely many independent vacuum solutions;
for instance, all plane waves (6) with diﬀerent null vec-
tors k
µ
are linearly independent. A general vacuum solu-
tion A
(vac)
µ
(x) is an arbitrary superposition of plane waves
with diﬀerent amplitudes and diﬀerent (null) wave vec-
tors k
µ
≡ (k, k),
A
(vac)
µ
(x) =

d
3
ke
ikµx
µ
a
k
, (9)
where a
k
are arbitrary k-dependent amplitudes. To ob-
tain a useful description of the space of vacuum solutions,
we need to somehow reduce this huge space to a ﬁnite-
dimensional vector space. Let us consider the values of
the vacuum solution A
(vac)
µ
(x
0
) at a ﬁxed spacetime point
x
0
≡ (t
0
, x
0
). The set of values A
(vac)
µ
(x
0
) for all possible
vacuum solutions A
(vac)
µ
comprises a vector space because
electrodynamics obeys the superposition principle. This
vector space is a subspace of the 4-dimensional vector
space of all possible 4-vectors A
µ
(t
0
, x
0
). Of course, the
dimension of this subspace does not depend on the cho-
sen point (t
0
, x
0
) because the equations of motion are
invariant under spacetime translations. This motivates
us to consider the following tentative deﬁnition.
Deﬁnition 2: A ﬁeld has n vacuum DOF if the val-
ues of all permissible vacuum solutions at a ﬁxed point
(t
0
, x
0
) comprise a vector space of dimension n.
In the next section we shall describe a more straightfor-
ward way to compute the number of DOF from this def-
inition. The key ingredient will be to perform a Fourier
transformation on the ﬁeld.
B. Computing the vacuum DOF via the Fourier
transform
The equations of motion (4) are linear diﬀerential equa-
tions with constant coeﬃcients, so their Fourier trans-
form can be written as
−k
ν
k
ν
˜
A
µ
(k) + k
µ
k
ν
˜
A
ν
(k) =
˜
j
µ
, (10)
3
where
˜
A
µ
(k) and
˜
j
µ
(k) are the four-dimensional Fourier
transforms of the 4-potential A
µ
(t, x) and the 4-current
j
µ
(t, x). The functions
˜
A
µ
(k) and
˜
j
µ
(k) are usually called
the Fourier modes of the ﬁeld and the current. The vac-
uum equations (with
˜
j
µ
= 0) can be written as

k
µ
k
ν
−k
2
δ
ν
µ

˜
A
ν
≡ W
ν
µ
˜
A
ν
= 0, (11)
where W
ν
µ
(k) is a “matrix” that depends on k
µ
and we
write k
2
≡ k
µ
k
µ
. It is clear that the solutions of the
homogeneous equation (11) comprise a linear subspace S
of the initial four-dimensional space of values of
˜
A
µ
, and
that the dimensionality of the subspace S will depend on
the 4-vector k
µ
. Therefore it will be convenient to analyze
the solutions at a ﬁxed k
µ
. In the physical space, these
solutions correspond to plane waves with a ﬁxed wave
vector k
µ
.
Note that
˜
A
µ
= λk
µ
(where λ is a scalar function of
k
µ
) is always a solution of Eq. (11), and thus the one-
dimensional subspace k
µ
spanned by the vector k
µ
is
always a subspace of S. But note that the gauge trans-
formations (2) are equivalent to
˜
A
µ

˜
A
µ
−ik
µ
˜ α. (12)
Since all solutions of the form

˜
A
µ
+ λk
µ

are gauge
equivalent, the space of physically relevant indepen-
dent solutions are found as a factorspace of S by the
one-dimensional subspace spanned by the vector k
µ
,
i.e. S
phys
= S/ k
µ
. The subspace k
µ
is thus the sub-
space of pure gauge solutions, i.e. solutions that are not
physically diﬀerent from
˜
A
µ
= 0.
To analyze the space S of solutions of Eq. (11) at a
ﬁxed k
µ
, it suﬃces to consider two cases: k
2
= 0 and
k
2
= 0. If k
2
= 0, Eq. (11) gives
˜
A
µ
= k
µ
˜
A
ν
k
ν
k
2
= ak
µ
, (13)
where a is some constant. However, this solution is en-
tirely within the gauge subspace and thus there are no
nontrivial vacuum solutions for k
2
= 0. In the remain-
ing case k
2
= 0 (but k
µ
= 0), we have
˜
A
µ
k
µ
= 0 and
thus
˜
A
µ
belongs to the three-dimensional subspace of 4-
vectors orthogonal to k
µ
. We have analyzed this subspace
above and found its basis in the form (k
µ
, l
µ
, m
µ
) with
the decomposition (7). In other words, S = k
µ
, l
µ
, m
µ

and therefore the factorspace S/ k
µ
= l
µ
, m
µ
is two-
dimensional.
Finally, we note that the mode with the zero wavenum-
ber k
µ
= 0 (called the zero mode) corresponds to the
homogeneous solution A
µ
(t, x) = A
(0)
µ
= const, i.e. the
4-potential A
µ
is constant in space and time. This solu-
tion can be reduced to A
µ
= 0 by a gauge transformation
of the form
A
µ
→A
µ
−∂
µ

A
(0)
ν
x
ν

(14)
and thus is not a physically nontrivial vacuum solution.
(We call a solution physically trivial if it is gauge equiv-
alent to the trivial solution A
µ
= 0.)
Our conclusion is that the space of vacuum solutions
for electrodynamics is two-dimensional. We have ob-
tained this result solely by analyzing the dimension of
a certain ﬁnite-dimensional vector space that we con-
structed as the factor space of the vacuum solutions by
the pure gauge subspace.
C. Vacuum DOF in general relativity
General relativity describes the dynamical geometry
of the spacetime through the metric g
µν
(x). We shall
consider the situation when the geometry is almost ﬂat
except for small gravitational perturbations, so that in
appropriate coordinate systems we have
g
µν
(x) = η
µν
+ h
µν
(x), |h
µν
| ≪1. (15)
The coordinate transformations
x
µ
→x
µ
+ ξ
µ
(x) (16)
play the role of gauge transformations in this theory. If
we consider only transformations with very small ξ, then
to ﬁrst order we may write the gauge transformation for
h
µν
as
h
µν
→h
µν
+ ξ
µ,ν
+ ξ
ν,µ
. (17)
Note that we do not need to employ covariant derivatives
such as ξ
µ;ν
because the diﬀerence between the covariant
and the ordinary derivatives is of second order in the
perturbations.
The symmetric tensor h
µν
has 10 independent com-
ponents, but not all of them are physically signiﬁcant,
and the number of independent components is further
reduced by the equations of motion. In this section we
shall compute the number of DOF in the gravitational
perturbations.
The vacuum equations of motion can be written as
G
µν
= 0. Since R
µν
= G
µν

1
2
Gg
µν
, we may also write
the vacuum equations in the form R
µν
= 0 which is a
little simpler. A calculation shows that the Ricci tensor
R
µν
is expressed through the metric perturbation h
µν
as
2R
µν
= η
αβ
(h
µα,νβ
+ h
να,µβ
) −h
µν
−h
,µν
, (18)
where = η
αβ

α

β
and h ≡ η
αβ
h
αβ
. It is convenient at
this point to perform a Fourier transform in all spacetime
coordinates and consider the Fourier images
˜
h
µν
and
˜
ξ
µ
.
Then the equation of motion R
µν
= 0 and the gauge
transformation (17) are rewritten as

˜
h
µα
k
ν
+
˜
h
να
k
µ

k
α
+ k
2
˜
h
µν
+
˜
hk
µ
k
ν
= 0, (19)
˜
h
µν

˜
h
µν
+ k
µ
˜
ξ
ν
+ k
ν
˜
ξ
µ
. (20)
4
We have thus obtained a linear, homogeneous equa-
tion for the 10 independent components of
˜
h
µν
tionally a gauge equivalence relation containing four free
parameters
˜
ξ
µ
. In principle we may now consider a 10-
dimensional linear space consisting of these independent
components Q
i
, i = 1, ..., 10, and rewrite the equation of
motion and the gauge equivalence relation in the matrix
form,
F
ij
Q
i
= 0, Q
i
→Q
i
+ G

ξ
µ
, (21)
where F
ij
and G

are suitable 10×10 and 10×4 matrices.
The solutions of the equation F
ij
Q
i
= 0 form a subspace
S in the 10-dimensional space; for a given solution Q
i
(0)

S, the solutions that are gauge equivalent to Q
i
(0)
diﬀer
by a vector from the pure gauge subspace G ⊂ S. The
problem of counting the DOF is then reduced to ﬁnding
the dimensions of these subspaces S and G ⊂ S, which is
a standard problem in ﬁnite-dimensional linear algebra.
The number of DOF is then dimS −dimG.
The calculations are thus made entirely straightfor-
ward but cumbersome. Rather than write the matrices
F
ij
and G

explicitly, we shall simplify the equations
for
˜
h
µν
in the tensor notation, use the gauge equivalence
relations, and arrive to the answer more easily.
First consider the 4-vector s
µ

˜
h
αµ
k
α
. Under a gauge
transformation, the vector s
µ
changes as
s
µ
→s
µ
+
˜
ξ
µ
k
2
+
˜
ξ
α
k
α
k
µ
, (22)
and the scalar product s
µ
k
µ
=
˜
h
µν
k
µ
k
ν
transforms as
s
µ
k
µ
→s
µ
k
µ
+ 2k
2
˜
ξ
α
k
α
. (23)
This transformation is trivial if k
2
= 0, so it appears
that we need to consider the cases k
2
= 0 and k
2
= 0
separately.
If k
2
= 0, then we can always ﬁnd a vector
˜
ξ
α
such
that
˜
ξ
α
k
α
= −
s
µ
k
µ
2k
2
, (24)
and then s
µ
k
µ
vanishes after the gauge transforma-
tion (23). After performing this gauge transformation,
we still have the freedom to perform additional gauge
transformations with vectors
˜
ξ
α
such that
˜
ξ
α
k
α
= 0. The
vector s
µ
would change under such a transformation as
s
µ
→s
µ
+ k
2
˜
ξ
µ
. (25)
Therefore we may set s
µ
= 0 by choosing
˜
ξ
µ
= −
s
µ
k
2
; (26)
note that this gauge transformation is allowed since we
µ
k
µ
= 0. After performing this last
gauge transformation, we have no more remaining gauge
freedom.
Using the relation s
µ

˜
h
µν
k
ν
= 0, we may simplify
the equation of motion (19) to
˜
hk
µ
k
ν
+ k
2
˜
h
µν
= 0. (27)
Computing the scalar product of Eq. (27) with k
µ
, we
ﬁnd
˜
hk
µ
k
2
= 0 and thus
˜
h = 0. Substituting this into
Eq. (27), we obtain k

h
µν
= 0 and ﬁnally
˜
h
µν
= 0. Thus
there are no physically nontrivial solutions of the vacuum
equations when k
2
= 0.
In the case k
2
= 0 (but k
α
= 0), we cannot perform
the gauge transformations as above and need to take a
diﬀerent approach. The trace
˜
h changes under a gauge
transformation as
˜
h →
˜
h + 2
˜
ξ
α
k
α
. (28)
Thus we may set
˜
h = 0 after a suitable gauge transfor-
mation. There will remain the freedom to perform gauge
transformations with vectors
˜
ξ
α
such that
˜
ξ
α
k
α
= 0.
After setting
˜
h = 0, we simplify the equation of mo-
tion (19) to
s
µ
k
ν
+ s
ν
k
µ
= 0, s
µ

˜
h
µν
k
ν
. (29)
Since the vector k
µ
= 0, the equation (29) has no
other solutions except s
µ
= 0. [To verify this, one ﬁrst
computes a scalar product with a vector a
µ
such that
a
µ
k
µ
= −1; the vector a
µ
exists as long as k
µ
= 0. Then
one obtains s
ν
= (s
µ
a
µ
) k
ν
and thus the vectors s
µ
and
k
µ
are parallel, s
µ
= λk
µ
. Then the substitution into
Eq. (29) gives λ = 0.]
Thus we have reduced the equations of motion and the
gauge equivalence relation to
˜
h ≡
˜
h
µν
η
µν
= 0,
˜
h
µν
k
ν
= 0, (30)
˜
h
µν

˜
h
µν
+ k
µ
˜
ξ
ν
+ k
ν
˜
ξ
µ
,
˜
ξ
µ
k
µ
= 0. (31)
At this point we could introduce an explicit basis in the
4-dimensional space and ﬁnd the remaining independent
components, but we shall instead proceed without choos-
ing a basis. The equations (30) represent 5 constraints
on the 10 independent components of
˜
h
µν
and thus we
can expect the space of solutions to be 5-dimensional.
The gauge equivalence relations are parametrized by the
vector
˜
ξ
µ
which is constrained to remain orthogonal to
k
µ
; such vectors form a 3-dimensional space. Thus it ap-
pears that there are 5 − 3 = 2 independent DOF. The
remaining task is to verify that the subspace of solutions
of Eq. (30) and the pure gauge subspace are indeed 5-
dimensional and 3-dimensional respectively.
The subspace S of solutions of the ﬁve constraints (30)
will have more than 10 − 5 = 5 dimensions if the con-
straints were not independent, i.e. if there exists a vanish-
ing linear combination of the constraints. This is equiv-
alent to the existence of ﬁve coeﬃcients A, B
µ
, not all
equal to zero, such that

µν
+ B
µ
k
ν
= 0. (32)
5
The scalar product of Eq. (32) with k
ν
yields A = 0 and
then the scalar product with any vector a
µ
such that
a
µ
k
µ
= 0 yields B
µ
= 0. Therefore the constraints (30)
are linearly independent and the space S has dimension
ﬁve.
Now we shall verify that the gauge equivalence reduces
this ﬁve-dimensional space to a two-dimensional factor
space. The gauge transformations (31) are compatible
with the constraints (30) in the sense that the gauge-
transformed solutions always satisfy the constraints.
Given a solution
˜
h
(0)
µν
of the constraints (30), we may
consider the set of gauge-transformed solutions
˜
h
(0)
µν
+
˜
ξ
µ
k
ν
+
˜
ξ
ν
k
µ
. (33)
This set is an (aﬃne) space and we would like to show
that this space has dimension three. The admissible vec-
tors
˜
ξ
µ
must be orthogonal to k
µ
and thus form a three-
dimensional space. So the space of gauge-transformed so-
lutions (33) cannot be more than three-dimensional, but
it might be less than three-dimensional if some gauge-
transformed solutions coincide with
˜
h
(0)
µν
. This however
cannot happen since the tensor
˜
ξ
µ
k
ν
+
˜
ξ
ν
k
µ
is never zero
for
˜
ξ
µ
= 0 and k
µ
= 0. Therefore the space of gauge-
equivalent solutions is always a three-dimensional aﬃne
subspace of S, and the remaining physically independent
space has dimension two.
Finally, we shall consider the last remaining case k
µ
=
0 (the zero mode). This perturbation mode corresponds
to the space-independent perturbation h
µν
(x) = h
(0)
µν
=
const. One can show that this perturbation is pure gauge,
i.e. is gauge-equivalent to the zero perturbation. For in-
stance, one can perform a gauge transformation
h
µν
→h
µν
+ ξ
µ,ν
+ ξ
ν,µ
(34)
with
ξ
µ
= −
1
2
h
(0)
µν
x
ν
, (35)
which will change h
µν
(x) = h
0
µν
into zero.
The conclusion is that small metric perturbations in
general relativity have two vacuum DOF and are limited
to modes with k
α
k
α
= 0, k
α
= 0.
This result was obtained using the equations of mo-
tion for metric perturbations h
µν
on a ﬂat background
η
µν
. A natural question is whether the number of vacuum
DOF depends on the background: for instance, whether
there are still two independent perturbation modes in
the neighborhood of a black hole. The answer is that the
number of vacuum DOF does not depend on the back-
ground.
To understand this issue, let us consider a ﬁxed space-
time point (t
0
, x
0
). We expect that perturbations have
the same number of DOF at all points. To describe the
behavior of the perturbations, it is suﬃcient to impose
the equations of motion in a small neigborhood of (t
0
, x
0
).
We may choose the coordinates in this neighborhood such
that the metric is approximately given by Eq. (15). Then
the reasoning shown in this section will apply to all modes
with suﬃciently small wavelengths, and we shall again
ﬁnd two vacuum DOF. The number of DOF is deﬁned as
the dimension of space of solutions restricted to the point
(t
0
, x
0
) and is independent of the chosen neighborhood
of that point (and of the choice of the point itself).
In the next section we shall generalize this construction
to an arbitrary nonlinear ﬁeld theory and show that the
number of vacuum DOF can be computed in essentially
the same way in all such theories.
D. Vacuum DOF in nonlinear ﬁeld theories
The deﬁnition of vacuum DOF and the calculation in
the previous sections relied on the fact that the Maxwell
equations are linear in the ﬁelds and that the solutions
form a vector space. For nonlinear ﬁeld theories (e.g. gen-
eral relativity) we therefore need to slightly modify our
deﬁnition of DOF and the corresponding computational
procedure.
Consider a ﬁeld theory with a ﬁeld Q
i
having N real-
valued components (i = 1, ..., N); for instance, in gen-
eral relativity the “ﬁeld” is the metric g
µν
(x) which has
10 independent components. We shall assume that all
independent components of all relevant ﬁelds are incor-
porated into the single “ﬁeld” Q
i
. Suppose that the ﬁeld
Q
i
(x) satisﬁes some nonlinear equations of motion of the
form
F

αβ
Q
i
, ∂
α
Q
i
, Q
i

= 0, (36)
where F(·, ·, ·) is some function. We have explicitly as-
sumed that the equations of motion are second-order in
derivatives of Q
i
, which is usually the case in ﬁeld the-
ories.
1
The ﬁeld theory may also admit gauge transfor-
mations of the form
Q
i
(x) →G
i
(∂
α
Q
i
(x), Q
i
(x), a(x)), (37)
where G
i
(·, ·, ·) is some function and a(x) is the param-
eter of the gauge transformation having an arbitrary de-
pendence on x. At this moment we do not consider a
speciﬁc form of the functions F and G and merely gen-
eralize the consideration of DOF to an arbitrary theory
of this form. Later we shall consider the vacuum DOF in
general relativity.
Using the analogy with electrodynamics where electro-
magnetic waves are small perturbations of the vacuum
A
µ
= 0, one can expect that the “vacuum degrees of free-
dom” in the ﬁeld theory (36) will be small perturbations
of a “vacuum state.” Thus we are lead to assuming that
1
Below we shall consider ﬁelds with higher-order derivatives as
well, but for now our considerations are valid only for ﬁelds with
second-order equations of motion.
6
Eq. (36) admits a special “vacuum state,” i.e. a solution
Q
i
vac
(x) which is physically interpreted as the “absence”
of the ﬁeld Q
i
. In electrodynamics the vacuum solution
is A
µ
= 0, but in a more general nonlinear ﬁeld theory
the vacuum solution may be a nontrivial function. For
instance, the vacuum solution in general relativity is the
ﬂat Minkowski metric g
µν
(x) = η
µν
.
Small perturbations δQ
i
(x) of the vacuum solution
Q
i
vac
(x) satisfy the linearized equation (36), namely
F
µνi
I

αβ
Q
i
vac
, ∂
α
Q
i
vac
, Q
i
vac

µν
δQ
i
+F
µi
II

αβ
Q
i
vac
, ∂
α
Q
i
vac
, Q
i
vac

µ
δQ
i
+F
i
III

αβ
Q
i
vac
, ∂
α
Q
i
vac
, Q
i
vac

δQ
i
= 0, (38)
where F
µνi
I
, F
µi
II
, and F
i
III
are the partial derivatives of
the function F(·, ·, ·) with respect to its ﬁrst, second, and
third arguments, and we have neglected terms of order
O

(δQ)
2

. The gauge transformation (37) induces the
corresponding transformation of the perturbation δQ
i
,
δQ
i
(x) →δQ
i
(x) +
∂G
i

α
Q
i
vac
, Q
i
vac
, a

∂a
a. (39)
However, we are now considering only inﬁnitesimal per-
turbations δQ
i
and so we cannot allow the vacuum solu-
tion Q
i
vac
(x) to be signiﬁcantly changed by a gauge trans-
formation. Thus we need to consider the gauge freedom
only with respect to inﬁnitesimal gauge transformations,
where the gauge parameter a(x) is of the same order as
the perturbation δQ
i
. So we may approximate a ≈ 0
in the arguments of G in Eq. (39) because the error is
second-order, and the gauge transformation of the per-
turbation δQ
i
can be written as
δQ
i
(x) →δQ
i
(x) + aG
i
III

α
Q
i
vac
, Q
i
vac
, a = 0

. (40)
Since Eq. (38) is linear in δQ
i
and Eq. (40) is linear in δQ
i
and a, the solutions form a linear space and we could now
apply the construction developed in the previous sections
to compute the number of the DOF, considering the small
perturbations δQ
i
as the “main ﬁeld.” Namely, the values
δQ
i
(t
0
, x
0
) at a ﬁxed spacetime point (t
0
, x
0
) the number
of DOF is equal to the dimension of the space S modulo
the gauge subspace.
To compute the number of DOF, we need to apply the
Fourier transform to Eqs. (38)-(40), but these are now
diﬀerential equations with nonconstant coeﬃcients such
as F
iαβ
I
(x). To overcome this diﬃculty, we note that the
number of DOF is determined only by the behavior of
the ﬁeld Q
i
in an arbitrarily small neighborhood of an
arbitrary point (t
0
, x
0
). The coeﬃcients F
iαβ
I
(x), etc.,
entering Eqs. (38)-(40), can be treated as constants in an
inﬁnitesimal neighborhood of (t
0
, x
0
), as long as the ﬁelds
are not singular at that point. Since the number of DOF
does not depend on the choice of the point (t
0
, x
0
), we
may choose (t
0
, x
0
) such that all coeﬃcients in Eqs. (38)-
(40) evaluated at that point are ﬁnite and nonsingular.
Thus we can replace the original system of equations (38)-
(40) by linear equations in δQ
i
with constant coeﬃcients,
evaluated at the point (t
0
, x
0
),
F
µνi
I
(x
0
)∂
µν
δQ
i
+ F
µi
II
(x
0
)∂
µ
δQ
i
+ F
i
III
(x
0
)δQ
i
= 0,
δQ
i
(x) →δQ
i
(x) + G
i
III
(x
0
)a.
The number of DOF in the ﬁeld δQ
i
can be determined
using this latter system of equations, since these equa-
tions describe a ﬁeld theory which is equivalent to the
original ﬁeld theory in an inﬁnitesimal neighborhood of
the point x
0
≡ (t
0
, x
0
). We now apply the Fourier trans-
form and arrive to the equations

−F
µνj
I
(x
0
)k
µ
k
ν
+ ik
µ
F
µj
II
(x
0
) + F
j
III
(x
0
)

δ
˜
Q
j
= 0,
(41)
δ
˜
Q
i
→δ
˜
Q
i
+ G
i
III
(x
0
)˜ a. (42)
These are now treated as ﬁnite-dimensional algebraic
equations depending on the parameter k
µ
and containing
an arbitrary scalar ˜ a. Since Eq. (41) is homogeneous, its
solutions form a subspace S in the original N-dimensional
ﬁeld space of vectors δ
˜
Q
i
. The gauge transformation (42)
generates a one-dimensional space P of “pure gauge” so-
lutions which is a subspace (possibly empty) within S.
The number of DOF is found as the dimension of the
space S modulo the pure gauge subspace, dim(S/P).
The consideration is straightforwardly adapted to the
case when the gauge transformation contains several
scalar parameters a
j
(x), j = 1, 2, ... instead of a(x). Note
that the dimensions of the spaces P and S depend on the
wave vector k
µ
and thus we need to enumerate all the pos-
sibilities. Typically these possibilities will depend only on
the value of k
µ
k
µ
; for instance, in electrodynamics and in
general relativity the dimension of P/S is nonzero only
for null vectors k
µ
.
III. SUMMARY